Harmonics compensation using feed forward and elman recurrent artificial neural networks

HARMONICS COMPENSATION USING FEED FORWARD AND ELMAN RECURRENT ARTIFICIAL NEURAL NETWORKS SUMMARY Keywords : Feed forward neural network, Elman's recurrent neural network, Fast and Standart backpropagation, Harmonics distortion, Active filter, Power quality In this study, the methods to apply the feed forward with two hidden layers and Elman's recurrent neural network for harmonic detection process in active filter are proposed. Two type neural networks with one and two hidden layers are used for this purpose. At the first step, we used three layers networks (input layer, hidden layer and output layer) with standart backpropagation and fast backpropagation learning algoritm. The hidden layer neurons and the output layer neurons use nonlinear sigmoid activation functions. In alternative networks, the output layer neurons use linear activation functions for comparison. At the second step, we used four layers networks (input layer, two hidden layers and output layer) with fast backpropagation learning algoritm. The hidden layer neurons and the output layer neurons use nonlinear sigmoid activation functions. For the training and test processes, input signals of the neural networks are the amplitudes of one period distorted wave. The amplitudes are taken 128 point at regular interval of time axis. The amplitudes are used to be input signals of the neural networks without any pre-processing. In order to make neural network enable to detect harmonics from distorted wave, it is necessary to use some representative distorted waves for learning. These distorted waves are made by mixing the component of the 5th, 7th, 11th, and 13 h harmonics in fundamental wave. For this purpose, 5th harmonic up to 70%, 7th harmonic up to 40%, 11th harmonic up to 10% and 13th harmonic up to 5% were used and approximately 2500 representative distorted waves were generated for training process. During the training process, the distorted waves were used for recognition. For the performance evaluation of the neural network structures, 5th harmonic up to 70%, 7th harmonic up to 40%, 11th harmonic up to 10% and 13th harmonic up to 5%, 17th harmonic up to 5%, 19th harmonic up to 2.5%, 23th harmonic up to 2.5%, 25th harmonic up to 2% were used and approximately 250 representative distorted waves were generated as a test set. After the training process is completed, the general distorted waves (test set) were used for recognition. As the result of recognitions at the training and test phase, output signal from each output unit means the content of each harmonic including the input distorted wave and these harmonics are eliminated from the distorted wave. As the result, the possibility of the feed forward and Elman's recurrent neural networks to detect harmonics is confirmed by compensating the distorted waves and it can be said that the feed forward and Elman's recurrent neural networks are effectively to be used for active filter. xxiv